Diagnostic accuracy of mammography readers and their memory performance have no correlation with each other
Kok, P., Pitman, A. G., Cawson, J. N., Gledhill, S., Kremer, S., Lawson, J., et al. (2010). Diagnostic accuracy of mammography readers and their memory performance have no correlation with each other. Journal of Medical Imaging and Radiation Oncology, 54(4), 315–324. doi:10.1111/j.1754-9485.2010.02177.x
Introduction: The study aims to determine if any association exists between visual memory performance and diagnostic accuracy performance in a group of radiologist mammogram readers.
Materials and Methods: One hundred proven mammograms (23 with cancers) were grouped into 5 sets of 20 cases, with sets being of equal difficulty. Pairs of sets were presented in 5 reads (40 cases per read, order random) to a panel of 8 radiologist readers (either present or past screening readers, with experience range from <1 year to>20 years). The readers were asked to either ‘clear’ or ‘call back’ cases depending on need for further workup, and at post-baseline reads to indicate whether each case was ‘new’ or ‘old’ (i.e. remembered from prior read). Two sets were presented only at baseline (40 cases per reader), and were used to calculate the reader's false recollection rate. Three sets were repeated post-baseline once or twice (100 cases per reader). Reading conditions were standardised.
Results: Memory performance differed markedly between readers. The number of correctly remembered cases (of 100 ‘old’ cases) had a median of 10.5 and range of 0–58. The observed number of false recollections (of 40 ‘totally new’ cases) had a median of 2 and range of 0–17. Diagnostic performance measures were mean (range): sensitivity 0.68 (0.54–0.81); specificity 0.82 (0.74–0.91); positive predictive value (PPV) 0.55 (0.50–0.65); negative predictive value (NPV) 0.89 (0.86–0.93) and accuracy 0.78 (0.76–0.83). Confidence intervals (CIs; 95%) for each reader overlapped for all the diagnostic parameters, indicating a lack of statistically significant difference between the readers at the 5% level. The most sensitive and the most specific reader showed a trend away from each other on sensitivity, specificity, NPV and PPV; their accuracies were 0.76 and 0.82, respectively, and their accuracy 95% CIs overlapped considerably. Correlation analysis by reader showed no association between observed memory performance and any of the diagnostic accuracy measures in our group of readers. In particular, there was no correlation between diagnostic accuracy and memory performance.
Conclusion: There was no association between visual memory performance and diagnostic accuracy as a screening mammographer in our group of eight representative readers. Whether a radiologist has a good or a bad visual memory for cases, and in particular mammograms, should not impact on his or her performance as a radiologist and mammogram reader.
peer-reviewed, diagnostic errors, mammography, memory, neoplasms